Building Library Recommendations with Big Data Graph Analysis

Over at NICS, Scott Gibson writes that University Libraries are looking to improve their recommendation systems based on the very same graph search technologies employed by companies like Amazon. Using the Nautilus supercomputer at Oak Ridge, Harriett Green, Kirk Hess, and Richard Hislop from the University of Illinois are data mining the circulatoin records of 14 million items in the UIUC Library.

Network Graph of Headings

Current search mechanisms for online library catalogs and digital collections are narrowed to searching by indexed subject terms, authors, titles, and selected key words,” Green says. “With such limited parameters, many materials — especially in collections as vast as the holdings at the University of Illinois Library — are rarely exposed in search results. We wanted to find a way to reveal these ‘underserved items’ and help users see the broadest selection of relevant resources in their searches. We sought to develop new methods of calculating relevancy and exposing the results, which ultimately we aim to incorporate into a recommender system for scholarly users.”

Using the open source Gephi application, the team was able visualize large network graphs, displaying the connections between headings as well as group them into communities by modularity scores, which will make their results more accessible to librarians who are unfamiliar with network theory.

Read the Full Story.

Resource Links: